Top AI Product

Most strong video models are closed APIs. LTX-2, from Lightricks, is the rare exception with truly open weights — it generates synchronized 4K video and audio together, up to 20-second clips at 50 FPS, and it runs on consumer GPUs rather than a rented cloud cluster.

## What LTX-2 is

The headline is that the sound is native, not bolted on. LTX-2 produces matching video and audio in one pass, with full access to weights, inference, and training code. That open posture is the whole pitch: studios and developers can run it locally, fine-tune it on their own footage, and avoid per-second API metering for generative video.

## The June 17 trainer

The fresh news is tooling. On June 17 Lightricks shipped a unified LTX Trainer that covers 13 training modes from a single YAML config — LoRA, IC-LoRA, and full fine-tuning across text-to-video, text-to-audio, image-to-video, video extension, inpainting, outpainting, audio-to-video, and video-to-audio. Packing that many conditioning paths into one trainer makes LTX-2 genuinely customizable, which is what separates an open model people actually build on from one they only demo once.